• Neil McGlohon's avatar
    ModelNet MPI Replay: Rand Perm Traffic Fix · daaf1739
    Neil McGlohon authored
    This commit changes the behavior of the random permuation traffic.
    My understanding is that s->gen_data is what keeps track of how much
    data has been generated by the LP where syn_data keeps track of how much
    data has been received by the LP. Since the random permutation traffic
    pattern cares about how much data has been _generated_, the use of
    syn_data in the gen_synthetic_tr() method's PERMUTATION case has been
    changed to gen_data.
    Additionally, the previous implementation of the prev_switch, as far
    as I could tell, was designed to help keep track of how much data had
    been generated by the LP since the last destination change. It didn't
    appear to do that. I've made a change that updates this field to be
    the amount of data generated at the time of the last switch, therefore
    the difference between gen_data and this field will equal the amount
    of data generated since last switch which is what we want to compare
    against the permuation threshold.
    In order to accomodate this behavior, I've had to add one more unsigned
    long into the nw_message ROSS message struct. The message size here is
    now 640 (changed from 624).
    Additionally, this source file hadn't yet been able to accomodate
    synthetic5 traffic option. It maxed out at 4. This fix was included
    in this commit as synthetic5 is the Random Permutation traffic
model-net-mpi-replay.c 89.8 KB